Reward AI showed humanoids climbing stairs, picking up clutter and unloading packages from a Cybertruck using its OM-1 model.
Reward AI’s demo described those behaviors as learned from under 30 minutes of human data recorded with sensor-equipped gloves, with no teleoperation—remote human control of the robot.
Why the data collection matters
Recording human activity with gloves instead of having someone remotely operate a robot could make collecting examples cheaper if it reduces the operator time needed to teach a task.
For practitioners, with no success rates supplied in the available evidence, the useful measure is human collection time per reliably learned task. If that drops to minutes, gathering examples for a new task could require less operator time—but repeatable performance is what would make that saving useful.